What kinds of problems does it handle?

Competition math problems in the style of IMO, Putnam, and USAMO.

What happens when it cannot solve a problem?

It uses calibrated abstention, meaning it says so rather than bluffing an answer.

math-olympiad

PackAI & models

math-olympiad is a plugin that lets an AI agent solve competition math problems such as IMO, Putnam, and USAMO questions. It checks the agent's proofs step by step using adversarial verification, where fresh-context verifiers attack proofs with specific failure patterns that self-verification misses. When a proof cannot be confirmed, it abstains rather than bluffing.

Unavailable. Delivery for this kind is on the roadmap — not serving yet.

Have an AI agent set up that can load plugins.

What your AI can do with it

Solves olympiad-style competition math problems (IMO, Putnam, USAMO)
Verifies proofs step by step
Uses fresh-context verifiers to attack proofs with specific failure patterns
Catches errors that self-verification misses
Abstains when a proof cannot be confirmed instead of bluffing

Getting started

Have an AI agent set up that can load plugins.
Add the math-olympiad plugin to the agent.
Give the agent a competition math problem and let it produce a proof.
Review the step-by-step verification results and any cases where the plugin abstained.

Signals

GitHub stars: 37k
Forks: 4k
Last commit: Sep 2026

Questions

What kinds of problems does it handle?: Competition math problems in the style of IMO, Putnam, and USAMO.
How does it check proofs?: Adversarial verification: fresh-context verifiers attack proofs with specific failure patterns, catching errors that self-verification misses.
What happens when it cannot solve a problem?: It uses calibrated abstention, meaning it says so rather than bluffing an answer.

Advanced

Item type: plugin
Key: anthropics-claude-plugins-official-math-olympiad
Source: github.com/anthropics/claude-plugins-official

github.com/anthropics/claude-plugins-official